Despite the increasing use of mobile technology, in particular smartphones, there are still situations where people in distress may not be able to use such devices. The authors describe a smart home with capabilities intended to assist frail but autonomous users in emergency situations, such as a fall.

This home uses existing home automation sensors augmented by an array of microphones to distinguish regular daily living activities from emergencies. To do so, an audio processing module needs to separate utterances expressing distress from daily living noises on the one hand, and spoken language as part of a conversation (for example, from a phone call) on the other. For this, the authors use a pragmatic approach, discarding utterances shorter than 150 milliseconds and longer than 2.2 seconds, based on an analysis of sample recordings. An intelligent controller analyzes the output of a speech recognition module within the context of the smart home environment, and initiates appropriate actions. In an experiment with 13 (healthy) individuals performing actions from four scenarios, an accuracy of 63.4 percent for activity recognition is reported.

Using the audio channel in combination with existing sensors in a home is a good choice for distress analysis, since speech is an interaction mode that is available even under severe circumstances. However, much work is left before the distress scenario can be resolved this way. So far, performance for a core task such as activity recognition is moderate, and no results are provided for speech recognition.